ABSTRACT

Background : Severe community acquired pneumonia (SCAP) is the most commomly encountered critical illness with high short-term mortality rate. Several general prediction models have been validated in predicting the outcome of critical illnesses, but still few evidences for SAPS 3 validation for SCAP. This study aims to evaluate the calibration and discrimination performance of SAPS 3 to predict the in-hospital mortality of SCAP.
M
ethods : We used retrospective cohort design design and collected the medical record of patients admitted with SCAP to emergency ward, high care and intensive care unit of Cipto Mangunkusumo Hospital from March 2019-March 2021). The 30 days in-hospital mortality was documented during the study. Collected data were analyzed using goodness-of-fit Hosmer-Lemeshow tset for the calibration performance and Receiver Operating Curve (ROC) for the discrimination performance of SAPS 3 toward the in-hospital mortality of SCAP.
Results :There were 484 subjects with SCAP with mortality rate of 49,2%. 73,8% was identified to have viral pneumonia due to severe/critical COVID-19 and 25,6% were bacterial. The calibration performance was good (p=0,519, correlation coefficient r=0,993 for the observed and expected from Hosmer-Lemeshow). Discrimination performance was excellent for total score with AUC 0,921 (95%CI 0,898-0,944). Calibration performance was also found to be very good in predicting mortality for general SAPS 3 formula (p=0,054, r=0,984) as well as for Austalasia’s (p=0,092, r=0,986). The probability of death prediction formula, however, both shows poor graphical goodness of fit for the calibration performance.
Conclusion :The calibration and discrimination performance of SAPS 3 score in predicting the in-hospital mortality of SCAP is good, but not for the general and Australasian formula in predicting mortality.
Keywords: Severe community acquired pneumonia, mortality, SAPS 3